what can be ground? noun type, constructions, and the ...bclevin/bls11slides.pdfmass-count...

Mass-Count DistinctionIntroduction to the Universal Grinder

ExperimentDiscussion

What Can Be Ground?Noun Type, Constructions, and the Universal

Grinder

Alex Djalali, David Clausen, Scott Grimm and Beth Levin

Stanford University

BLS, February 12, 2011

Djalali et al. Noun Type, Constructions, and the Universal Grinder



Mass-Count Distinction

Introduction to the Universal Grinder

Experiment

Discussion




Background: The Mass-Count Distinction

What does it mean for a noun to be morphosyntactically mass orcount?





Count nouns (dog, chair):

I permit plural marking (dogs, chairs)I modification by cardinal quantifiers (one dog/chair)I may allow modification by determiners implicating plurality

(many dogs, several chairs)





Mass nouns (sand, tar, water):

I do not permit plural marking (*tars, *sands)I nor cardinal quantifiers or determiners implicating plurality

except on kind interpretations (*one tar, *several sands)I may allow modification by much (much sand)




Current approaches

Mass or count is a property of nouns, not extensions.

I Inherency ⇒ Choice is predetermined by the nature of theentity named

I Arbitrariness ⇒ Choice is not-predetermined, though theremay be some regularities or tendencies in lexicalization asmass or count




Current approaches

Two types of evidence used to support (versions of) thearbitrariness position:

I doublets:

As regards the semantic distinction: it seems thatthere is nothing in the referent of the terms thatshould make fruit mass and vegetable count, baklavamass and brownie count, rice mass and bean count.(Pelletier 1991: 497)

I shiftability of noun meaning with respect to countabilitystatus (focus of this talk)

There is apple in the salad.




Introduction to the Universal Grinder

After the grenade has exploded in the enemy bunker,Rambo might enter and notice that the walls containthree different muds mixed with sm. soldier.(Pelletier 1991: 497)




Universal grinder

Consider the ‘Universal Grinder’, a device that takes in anobject corresponding to the count term and spews outthe finely ground matter of which it is made. A hat, forinstance is fed into it and afterwards there is hat all overthe floor. This is so despite the fact that there is anotherword we might have used (for example, felt or straw).So for any word one would wish to call a count term,there is a related mass term designating, roughly, thestuff of which it is made. (Pelletier 1991: 497)




Universal grinder

I Universal grinder hypothesis ⇒ Every count noun can havea mass interpretation

(1) There is dog all over the highway.

(2) There is oil all over the highway.




Universal packager

I Universal packager ⇒ The ‘inverse’ operation, which resultsin count interpretations for typically mass nouns

(3) Three beers please. [= three servings of beer]




Universal grinder and packager: Implications

Universal grinder and packager data are often taken as evidence forthe arbitrariness of the mass-count distinction:

A noun’s status is not tied to the lexical item itself but isnecessarily computed at the NP level (Allan 1980, Bunt 1985).




Non-universality of universal grinder/packager

If the effects of the grinder and packager were truly universal, theyshould apply uniformly across all nouns, but these operations arerestricted.




Non-universality of universal packager

The packager is restricted:

I Packaging largely occurs with nouns whose referents arealready associated with conventionalized units of packaging

(4) Three beers please. [= three cans of beer; 6= three kinds ofbeer]

(5) #Rices adorn the altar.

(6) #I’ll have a dirt here. (6= I’ll have a shovelful of dirt here)(Filip 1999: 62)

Reason: Conventionalization plays a major role in its successfulapplication.




Non-universality of universal grinder

The grinder is restricted:

I It is difficult to grind highly individual objects, especiallyartifacts (Chierchia 2010: 106):

(7) There is dog all over the highway.

(8) #There is mug/toaster on the table.

(9) #Would you care for some more pea? (Fillmore 1989: 49)





I Measure Terms: It is impossible to grind *du kilo (‘somekilo’), *de la categorie (‘some category’) and *du chapitre(‘some chapter’) in French (Galmiche 1989: 68).

I Domain Particular Restrictions:

“The grinding function in English does not generally apply tothe names of plants to derive the names of cooking oils, but itdoes apply to derive the names of oils and essences inperfume:

(10) ?We fried the chicken in safflower (olive, corn, etc.)

(11) The lotion contains lavender (ylang-ylang, jasmine,bergamot)” (Nunberg and Zaenen 1990: 389)





Ground interpretations are not observed in the expected syntacticcontexts in Chinese (Cheng, Sybesma and Doetjes 2008):

(12) a. qiang-shangwall-top

douall

shıcop

gou.dog

There are dogs all over the wall.NOT: There is dog all over the wall.

b. qiang-shangwall-top

douall

shıcop

gou-rou.dog-flesh/meat

There is dog(meat) all over the wall.

c. dı-shangfloor-top

douall

shıcop

shuı.water

There is water all over the floor.





I The inability to grind cannot be attributed to the availabilityof classifiers

I Similar outcomes in other languages disposing of a baresingular, cf. Brazilian Portuguese (N. Silveira p.c.)

(13) Temhave

cachorrodog

nain-the

estradaroad

todaall

#There was dog all over the road.




Questions for study

I Are grinding restrictions tied to different noun classes?

I Are grinding restrictions tied to different constructions?




Experimental Evidence

Experiment: Question 1

Question 1: Does grinding acceptability vary across natural nounclasses?

Hypothesis 1: Nouns will systematically vary by class inacceptability when ground.

Low Acceptability ⇒ High Acceptability

Artifacts vs. Natural KindsComplex Artifacts vs. Simple Artifacts

Groups vs. Individuals





Experiment: Question 2

Question 2: Does grinding acceptability vary according to thenature of the situations depicted?

Hypothesis 2: Nouns will systematically vary in acceptabilityacross situations.

Low Acceptability ⇒ High Acceptability

Obscure Situations vs. Conventional Situations

(14) There was squirrel all over the road.

(15) There was blueberry all over the road.





What constructions should be used in the experiment?

Grinding constructions are usually illustrated with ‘paste’constructions (Borer p.c.):

I There is X all over the Y .

I There is X in the Y .

I A Y eats X .





Are we missing other grinding constructions?

Other constructions, such as comparative constructions, arepurported to allow grinder interpretations:

(16) Hire more car for less money.

(17) Chevy Volt is more car than electric (ad, 2010)

These comparisons tacitly make reference to the number ofproperties a car has, for example, rather than the amount of ‘stuff’of which that car is made.





Are we missing other grinding constructions?

Mass terms used in such comparative contexts show a similarinterpretation.

(18) This brand is more whiskey than that one.

Here, the referent of this brand has more properties relevant tobeing ‘whiskey’ than the referent of that one does.

Conclusion: Such comparatives do not qualify as grindingconstructions.





The constructions used in this study

In our experiment, we will restrict our attention to ‘paste’constructions.





Experimental Design

I Ask subjects to rate the acceptability of nouns of varioustypes in several grinding constructions.

I Each construction is presented in two instantiations to depictdistinct situations (contexts).





Materials

210 total stimuli

7 noun types; 5 tokens of each

I shape: tube, cylinder, sphere, cone, cube

I group terms: forest, bouquet, fleet, swarm, committee

I members of group terms: tree, flower, ship, bee, person

I simplex artifacts: hammer, towel, shirt, bucket, pencil

I complex artifacts: toaster, car, computer, violin, forklift

I animals: squirrel, snake, robin, butterfly, pig

I food stuff: steak, apple, cracker, yam, pea





Materials

3 grinder constructions; 2 contexts each

There is NOUN all over the floorhighway

There is NOUN in the breadconcrete

A robot eats NOUNA termite





Methods

295 subjects rated Universal Grinder sentences for acceptabilityusing a 1 (unacceptable) – 7 (acceptable) value Lickert scale.

Each subject rated a counterbalanced set of 7 target sentences and8 filler sentences in random order.

Judgments collected using Amazon Mechanical Turk.





Methods

Instructions:

For each sentence, please mark on a scale of (1) - (7)how acceptable you think it is. A rating of (1) meansthat you think the sentence is not an acceptable Englishsentence at all. A rating of (7) means that you think thesentence is a perfectly acceptable sentence in English.Rating in between these scores indicate a gradientinterpretation of acceptability.





Presentation





Results: Overview

2065 total ratings collected.

Average time of 4.8 seconds per rating.

Average of 10 ratings per stimulus.





Results: Universal Grinder Sentences

Low average acceptability ratings for Universal Grinder sentences:

Type Mean SD p < .05Grinder 2.331 1.806

Filler 5.679 1.848 *





Results: Noun Types

Plot of Means

Noun Groups

Rat

ing

●

● ●●

●

●

●

2.0

2.5

3.0

3.5

GROUP COMP SHAPE SIMP IND ANIM FOOD

F: 19.99 on 7 and 1859 DF, p-value: < 2.2e-16





Results: Noun TypesLow to High Acceptability

group termscomplex artifactsshapesimplex artifactsindividual group members

< animals < foodstuff





Results: Noun TypesCoefficient Estimates

Group Coef SE Pr > tGroup -0.21 0.06 ***

Complex -0.16 0.06 **Shape -0.16 0.06 **Simple -0.13 0.06 *

Individual -0.06 0.06Animal .08 0.06Food .60 0.06 ***





Results: NounsSignificant across type variation differences

Plot of Means

Nouns by Group

Rat

ing

●

●●

●

●

●

●

●

●

● ●

●

●

●●

● ●● ●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

1.5

2.0

2.5

3.0

3.5

4.0

4.5

forest toaster forklift cube shirt tree bee robin apple pea

Levene’s Test Df: 6 F: 11.454 Pr > F: 1.296e-12





Results: ContextNo significant effect of context

Plot of Means

Contexts

Rat

ing

●

●

●●

●

●2.

12.

22.

32.

42.

52.

62.

7

ROBOT TERMITE BREAD HIGHWAY FLOOR CONCRETE

F-statistic: 1.504 on 6 and 1860 DF, p-value: 0.1728





Results: ConstructionsNo significant effect of construction

Plot of Means

Construction

Rat

ing

●

●

●2.

22.

32.

42.

5

EATS OVER IN

F-statistic: 2.221 on 3 and 1863 DF, p-value: 0.08377




Summary of Results

I Most noun types were rejected by our participants.

I But foodstuff and animals were more felicitous.

I There was not a significant effect of grinding constructions.




Significance of Results

I The higher acceptability of grinder sentences with foodstuffand animals may be due to their conventional associations:

I dual life of food nouns as natural entities or processed foodstuff

I animals as natural entities or their flesh




Experimental results align with naturally occurring uses

Only mass nouns are attested in naturally occurring uses ofgrinding constructions/contexts.

Evidence: The results of a series of Google searches for instancesof these constructions.

The question: What can fill the X position in ‘paste’ contexts:

I There is X all over the wall

Expectation: If grinding is allowed, then there should be attestedexamples where X is filled by a basically ‘count’ noun.




Experimental results align with naturally occurring uses

The results: The X position is overwhelmingly filled by nounsdesignating liquids/mush or granular aggregates—that is,prototypical mass nouns:

I There is X all over the wall

I liquids/mush: blood, water (50%)

I granular aggregates: sugar, salt (31%)

I other: meat, love (19%)

I No instances of true grinder readings

Thus these contexts have an affinity for ‘mass’ notions; but . . .

Key Point: Positive evidence for speakers using grindingconstructions to grind something is difficult to come by.




Conclusion

I The grinder is not universal.

I Noun referents may be ground, but the success of thisoperation is dependent on the noun type.

I These results argue against the arbitrariness approach to themass-count distinction.




Thanks

Thanks to the audiences of the Workshop on Empirical,Theoretical and Computational Approaches to Countability inNatural Language, Bochum, Germany and the StanfordPsychology of Language Tea, Sven Lauer, Tania Rojas-Esponda aswell as to Mario Villaplana for his assistance in data collection.


what can be ground? noun type, constructions, and the ...bclevin/bls11slides.pdfmass-count...

Documents